An experiment in teaching a language without relying on translation • Hacker News Board

Re: An experiment in teaching a language without relying on translation

#71

Post by gibba999 » Mon, Jan 25, 2021, 11:15 AM UTC

henrikschroder wrote: post #34

tsimionescu wrote: post #19

Earlier quoted context omitted.

I think that one problem that eventually happens with this approach, which is a problem I had with Duolingo as well, is that very soon after learning a few words you need to start learning basic grammar. In these approaches that don't explain the grammar at all, you need to spend effort pattern matching to pick up the right grammatical rules to construct your own sentences, and you are never sure that you're doing th…

Just one more anecdote, but this is what turned me off from duolingo completely. You can't just learn words or sentences alone, you have to have at least some grammar in there. You need lists of prepositions and pronouns, and you need a little bit of at least simple rules for verbs and nouns. Yes, it's boring but otherwise you'll really struggle.

I don't find Duo helpful in isolation for this (and many other) reasons.

But I've found ~15 minutes per day to be fairly helpful in conjunction with other activities.

Re: An experiment in teaching a language without relying on translation

#72

Post by AnIdiotOnTheNet » Mon, Jan 25, 2021, 1:19 PM UTC

imtringued wrote: post #57

AnIdiotOnTheNet wrote: post #13

Earlier quoted context omitted.

Well the good news is that the kana are really low hanging fruit. You can learn them in a day and as long as you continue learning Japanese you will be constantly exposed to them. There are only just over a hundred of them and, unlike English, they are only ever pronounced one way[0]. Kanji are significantly more effort, with 2k characters to learn and multiple pronunciations for each. Interestingly, while you will p…

It took me at least a month to learn all Hiragana + katakana using getkana and they only really stuck after I began learning vocabulary. Learning them in a day is impossible. You can look them up if you want to get started on day one but that is completely different from actually knowing the by heart.

Like I said, if you continue studying Japanese you'll be continually exposed to them which will make them stick.

By "learn them in a day" I meant one day of decently focused effort to memorize the syllable for each kana character. Between flashcards, pronunciation, and writing it is very doable because that's what I did.

Re: An experiment in teaching a language without relying on translation

#73

Post by timw4mail » Mon, Jan 25, 2021, 2:15 PM UTC

naniwaduni wrote: post #48

_solr wrote: post #15

Hi I am the OP. I just discovered that my submission has made it to the front page. A quick word on the rationale for not using translation. All of the classic apps (Anki, Duolingo, Memrise, ...) are to some extent translation based. Translating has three scalability issues. First, if you want to be inclusive and your product to be used by everyone you have to translate the whole product in a long list of languages.…

Or worse—僕はうなぎだ!

You're an eel?!

Re: An experiment in teaching a language without relying on translation

#74

Post by cyphar » Mon, Jan 25, 2021, 4:00 PM UTC

jjeaff wrote: post #49

Swizec wrote: post #33

Earlier quoted context omitted.

> In these approaches that don't explain the grammar at all, you need to spend effort pattern matching to pick up the right grammatical rules to construct your own sentences, and you are never sure that you're doing the right thing. To be fair, you learn your first language by inference and don't start learning the rules until you're some 10 years into fluency. Even as late as high school we were taught grammatical r…

We may not teach a codified grammar rule set until later, but we are definitely learning grammer as we learn language for the first time. My 2 year old will say things like, "I like she". After which we will correct her, "I like her". She has never heard anyone say "I like she" but she is re-ordering and inventing her own sentences using unspoken rules. (She may have generated that sentence from hearing "where is she…

There is a fair amount of evidence that corrections (negative evidence) do not actually improve the grammar of children who are still developing their language skills[1], for two reasons: adults are bad at correcting mistakes consistently, and children do not internalise corrections because language acquisition is not a conscious process.

In addition, even in early development children actually do appear to have pretty strong understanding of their native language's syntax[2] even if they cannot produce adult-level English. There are certain classes of mistakes they do not make (indicating they have internalised many more rules than may be apparent), and they can understand sentences which have grammar that they do not produce themselves (meaning that the level of output is not a good indication of their language development).

[1]: https://www.youtube.com/watch?v=a7Un06tDOn0 [2]: https://www.youtube.com/watch?v=zmghbKNiI1k

Re: An experiment in teaching a language without relying on translation

#75

Post by cyphar » Mon, Jan 25, 2021, 4:02 PM UTC

tsimionescu wrote: post #64

Xenograph wrote: post #40

Earlier quoted context omitted.

Not that I disagree necessarily, but to play devil's advocate: kids don't learn any formal grammar rules before they speak their first language fluently.

Sure, but there are two caveats: 1. Kids take about 7-10 years of constant immersion to become fluent in a language 2. Child brains are different from adult brains, and it is likely that the ability to pick up a language in this way degrades significantly after a certain age I'm not saying it is impossible to pick up a language without being told the grammar rules, just that it's much faster if you do learn them.

> 2. Child brains are different from adult brains, and it is likely that the ability to pick up a language in this way degrades significantly after a certain age.

There is plenty of evidence which counters the latter part of this claim (see Stephen Krashen and the input hypothesis).

Re: An experiment in teaching a language without relying on translation

#76

Post by Kwpolska » Mon, Jan 25, 2021, 8:40 PM UTC

knolax wrote: post #70

gryson wrote: post #12

But... why? Translation is not something to be avoided for the adult language learner. We should use every advantage at our disposal. You can spend hours trying to guess the meaning of a word through pictures and context (and not getting anywhere far), or you can spend a minute reading the translated meaning and then spend hours understanding it in context (and reinforcing it, and learning its semantics). I wonder if…

> Translation is not something to be avoided for the adult language learner. We should use every advantage at our disposal. Because you're relying on the translator to accurately convey the meaning of words to you, which translators rarely do. Most professional translators and language instructors don't exactly have the best grasp of language themselves. They do some 1 to 1 translation of "monkey" and 猿 as if the con…

The Japanese Wikipedia is not a translation of the English Wikpedia. Wikipedia isn’t a good place to learn a language, and especially not by comparison. For example, here’s the Polish Wikipedia page:

404 Not Found

Oh, and here’s the German Wikipedia page:

404 Not Found

…hold on, that’s not right. Both Wikipedias have a disambiguation page, and the first link on them goes to their equivalent of Simian. The point is, every Wikipedia has its own structure, its own rules of what should be a page, and articles might be based on sources in that language, not in English.

Re: An experiment in teaching a language without relying on translation

#77

Post by ralferoo » Tue, Jan 26, 2021, 10:12 AM UTC

So, I'd like to try this and I think it's a viable approach... But a language, especially one with a different script that I can't read, really needs some audio clues too, because that's a big thing that'll help to be able to internalise a word.

I mean I can guess banana from the picture and the repeated na sound, but I have absolutely no way of internalising the other characters.

I know of the existence of katakana and hiragana, but I don't actually know any characters, and if you intend this to be teaching the language, you need some way of teaching the basic building block you're relying on.

Also, as I'm learning Chinese, the kanji aren't really a problem and actually I can guess the meaning, but even then I can't internalise the kanji in Japanese with no idea what it's supposed to sound like. At best, it'll just be Chinese plus squiggles. To me it'd just be "yuan mmm banana mmm shi mmm mmm".

If the goal is only to read, then I guess eventually you might start to recognise some of the grammar characters. But if the goal is to learn the language, and presumably to speak it to, then I don't see how this can succeed. Unless of course, I'm missing something obvious and there is supposed to be audio...

Re: An experiment in teaching a language without relying on translation

#78

Post by toto444 » Tue, Jan 26, 2021, 1:41 PM UTC

ralferoo wrote: post #77

So, I'd like to try this and I think it's a viable approach... But a language, especially one with a different script that I can't read, really needs some audio clues too, because that's a big thing that'll help to be able to internalise a word. I mean I can guess banana from the picture and the repeated na sound, but I have absolutely no way of internalising the other characters. I know of the existence of katakana…

> Unless of course, I'm missing something obvious and there is supposed to be audio...

Learning Hiragana and Katakana as well as their pronunciation is a pre-requisite. That was mentioned before but for some reason I removed it.

Re: An experiment in teaching a language without relying on translation

#79

Post by jjeaff » Wed, Jan 27, 2021, 10:13 PM UTC

cyphar wrote: post #74

jjeaff wrote: post #49

Earlier quoted context omitted.

We may not teach a codified grammar rule set until later, but we are definitely learning grammer as we learn language for the first time. My 2 year old will say things like, "I like she". After which we will correct her, "I like her". She has never heard anyone say "I like she" but she is re-ordering and inventing her own sentences using unspoken rules. (She may have generated that sentence from hearing "where is she…

There is a fair amount of evidence that corrections (negative evidence) do not actually improve the grammar of children who are still developing their language skills[1], for two reasons: adults are bad at correcting mistakes consistently, and children do not internalise corrections because language acquisition is not a conscious process. In addition, even in early development children actually do appear to have pret…

Interesting video. But it seems that the definition of negative evidence is telling the child "no" and then correcting them.

When I say we "correct her", I simply mean that we repeat back the sentence correctly. It seems to work well as she will often repeat it back again how we say it or sometimes just continue on. We never withhold understanding because the grammar is wrong. I can definitely see how the example given of correcting with "no, that's not proper grammar" would not be effective with young ones.

By repeating it back correctly, she is given another example of correct grammar and I know of no other way for a young child to learn than to hear it spoken more.

Re: An experiment in teaching a language without relying on translation

#80

Post by cyphar » Thu, Jan 28, 2021, 3:41 AM UTC

jjeaff wrote: post #79

cyphar wrote: post #74

Earlier quoted context omitted.

There is a fair amount of evidence that corrections (negative evidence) do not actually improve the grammar of children who are still developing their language skills[1], for two reasons: adults are bad at correcting mistakes consistently, and children do not internalise corrections because language acquisition is not a conscious process. In addition, even in early development children actually do appear to have pret…

Interesting video. But it seems that the definition of negative evidence is telling the child "no" and then correcting them. When I say we "correct her", I simply mean that we repeat back the sentence correctly. It seems to work well as she will often repeat it back again how we say it or sometimes just continue on. We never withhold understanding because the grammar is wrong. I can definitely see how the example giv…

> But it seems that the definition of negative evidence is telling the child "no" and then correcting them.

The video goes through several examples from peer-reviewed studies, and several of them involve "repeating back the correct sentence". In fact while the video did mention the fact that saying "no" just makes communication harder, the majority of the video is explaining why giving corrected examples of grammar doesn't necessarily help.

The video talks about your approach specifically here[1].

> I know of no other way for a young child to learn than to hear it spoken more.

That is true, but repeating back a specific sentence is not necessarily the most effective manner of helping them acquire an aspect of grammar. Language acquisition isn't a conscious process, so focusing on a mistake doesn't help -- and repeating a sentence with correct grammar is only a single example of that particular grammar concept, which the child still needs to hear many more times before they actually acquire.

It should also be noted that one of the studies (looking at what sentences parents would correct) found that "there is not even a shred of evidence that approval and disapproval are contingent on syntactic correctness". So it's quite likely that you aren't actually correcting grammatical mistakes -- most of the time you're probably correcting sentences with incorrect meaning and pronunciation.

I mean, it probably doesn't hurt or anything, it's just probably not as helpful as you might think it is.

[1]: https://youtu.be/a7Un06tDOn0?t=272